Easily control traffic between Apps and LLM using data security policies

Integrate quickly using existing openai compatible chat completions api by just changing base url

import openai

api_key = "YOUR_OPENAI_KEY"
client = openai.OpenAI(api_key=api_key,base_url="https://dev.defendai.tech")
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Is this valid SSN 123-45-6789 ?"}
    ],
    stream=False
)
print(response.choices[0].message["content"])

Easily Add policies to Block/Alert/Anonymize Prompts or Responses by App and LLM Model

from defendai_wozway import Wozway
import os
api_key = "YOUR_OPENAI_KEY"
client = Wozway.Client(api_key=api_key,base_url="https://dev.defendai.tech")
policies = [
    {
        "policy_type": "Compliance",
        "direction": "PROMPT",
        "attributes": ["PII", "GDPR"],
        "action": "ALERT",
        "model": "gpt-4o",
        "app": "CopilotApp"
    },
    {
        "policy_type": "Security",
        "direction": "PROMPT",
        "attributes": ["Malware", "Phishing"],
        "action": "BLOCK",
        "model": "deepseek-r1",
        "app": "Global"
    },
    {
        "policy_type": "Privacy",
        "direction": "RESPONSE",
        "attributes": ["SSN"],
        "action": "ANONYMIZE",
        "model": "claude-sonet-3.7",
        "app": "StockAgent"
    }
]
client.apply(policies)

This service allows developers to easily secure requests and responses between their chat apps and LLM cloud services like OpenAI, Groq, Gemini , Anthropic , Perplexity and more using policies through cloud driven UI or APIs available via the wozway sdk

Table of Contents

• Introduction
• Features
• Architecture
• Prerequisites
• Installation
• Usage
• Developer SDK
• Provision Tenant Video
• UI Walkthrough Video
• Support
• License

Introduction

Wozway by DefendAI provides a secure and efficient way to manage communication between your chat client applications and various LLM cloud services. By running a simple script, you can start a tenant that proxies your LLM requests and response through a policy driven llm proxy , adding an extra layer of security and abstraction. Each tenant gets a cloud domain with ability to control traffic through the service

Features

• Secure proxying of requests to Multiple LLM Providers (OpenAI, Groq, etc.)
• Policy based enforcement using API or UI
• Support for Regex , Anonymization , Code , Phishing URL , Prompt and Response detection
• Similarity search with millions of malicious prompts database to block/alery admin
• Incident , Activities and Dashboard for monitoring usage across all Gen AI Apps

Architecture

Wozway runs alongside client apps and intelligently re-routes only LLM bound traffic to DefendAI Cloud for policy enforcement and subsequent forward to LLM Providers like Open AI , Anthropic , Groq etc .. All other traffic is routed without any interference to its intended destination .. Wozway can intercept both prompt and response for running full security check

Prerequisites

• Python 3.7+
• pip package manager
• API keys for the LLM services you intend to use (e.g., OpenAI or GROQ API key)
• Docker

Installation

1. Clone the repository

   git clone https://github.com/Defend-AI-Tech-Inc/wozway.git
   cd wozway

2. Install dependencies

   pip install -r requirements.txt

Usage

Starting the Tenant

Run the following script to start your tenant:

python start_tenant.py

This script will ask few questions to setup the cloud account and domain . It will start local version of openwebui integrated with wozway to proxy all requests throught the cloud proxy to the LLM Provider . Prompt and Responses are intercepted based on policies configured in cloud console

SDK Installation

The SDK can be installed with either pip package manager.

PIP

pip install defendai-wozway

SDK Example Usage

Simple Getting Started Example

from defendai_wozway import Wozway
import os

with Wozway(
    bearer_auth=os.getenv("WOZWAY_BEARER_AUTH", ""),
) as s:
    res = s.activities.get_activities()
    if res is not None:
       for activity in res:
          print(activity["prompt"])
      pass

Detailed SDK documentation can be found here SDK Documentation

Installation Video

UI Video Walkthrough

defendai_overview.mp4

Support

If you encounter any issues or have questions, please open an issue on the GitHub repository or contact our support team at support@defendai.tech.

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.